Developers using answers API can access the same model that has strength Chat GPT Looking: GPT-4O Search and GPT-4O Money Find. This model can Browse the web Responding to questions and referring to sources in response.
This is noteworthy because Openi says the web search capacity dramatically improves the accuracy of the facts of its AI models. On the open Simple Saka BenchmarkWhich aims to measure the Conflicts rate, GPT -4 Ou Search scored 90 %, while GPT -4 and Money Search achieved 88 % -both performed a large GPT -4.5 model without any search, which scored 63 %.
Despite these improvements, technology still has significant limits. Except for CUA cases that websites properly navigated, the better capacity of the search does not fully solve the problem of AI Confusion, GPT -4Os still make 10 % of mistakes.
Along with API’s answers, Openai issued an open source Agents sdkProviding developers, providing free tools to connect models with internal systems, enforce safety measures, and monitor agent activities. This toolkit is after the Openi’s pre -release SheepA framework for arcketing several agents.
These are still early days in the field of AI agent, and things will likely improve rapidly. However, at this time, the AI agent’s movement is at risk of non -real claims, as it appeared earlier this week when Discovered users They have a Chinese start -up butterfly effect Monos you have The agent platform failed to fulfill many of its promises, highlighting the permanent difference between promotional claims and practical functionality in the emerging technology category.